Recognition of aluminum electrolysis overheat trend based on DA-LSTM Neural Network

Abstract

Superheat is the difference between the temperature of electrolyte and the temperature of primary crystal in aluminum electrolysis production, which is related to the physical field, current efficiency and electrolytic cell life and other important indicators in production. Therefore, by monitoring and identifying the degree of superheat, various parameters and blanking in the aluminum electrolysis process can be reasonably adjusted to keep the degree of superheat within a reasonable and stable range, which is of great significance to the efficient operation of the entire aluminum electrolysis cell. At present, many scholars have studied the identification of superheat and achieved a certain accuracy, but there are stiff few studies on the identification of the trend of superheat change. Therefore, in this paper, by mining the time sequence information of various data in the production process of aluminum electrolysis, the Long Short Term Memory (LSTM) algorithm with dual-stage attention mechanism (DA-LSTM) is used to classify and identify the superheat trend. The first stage of DA-LSTM introduces input feature attention to increase the weight of more relevant features. In the second stage, time step attention is introduced, and different time steps are weighted. Finally, the effectiveness of this method is verified by comparing with other methods, and it has higher accuracy.